Artificial Analysis Drops First Coding Agent Benchmark — Composite Score Across SWE-Bench-Pro, Terminal-Bench, SWE-Atlas

X/Twitter·medium signal

Artificial Analysis released the first comprehensive coding agent benchmark, measuring performance, cost, and execution time across end-to-end software engineering tasks with a composite pass@1 score spanning SWE-Bench-Pro-Hard-AA, Terminal-Bench v2, and SWE-Atlas-QnA. Early leaderboard results show Boba by stealth leading with arena score 1103, followed closely by Claude Sonnet 4.6 (1101) and Claude Opus 4.7 (1035). For builders evaluating which agent to integrate, this is the first apples-to-apples comparison that goes beyond model benchmarks to measure actual agent performance.